Re: XPath Data Model proposal

At 02:38 PM 6/5/2003 -0400, John Cowan wrote:

>Jonathan Robie scripsit:
>
> > To me, the Infoset-only mapping in our document needs to be consistent 
> with
> > the standard definition of XML. There are many ways we could decide to
> > doctor an Infoset or reinterpret it, and I think this is a dangerous path
> > to go down for mappings defined in our specifications.
>
>Well, if I understand Norm correctly, xsi:nil is to be interpreted in
>Infoset-only processing in any event, and it doesn't seem to me too much
>of a stretch to require the interpretation of xsi:type as well.

Hmmm.

If I understand Norm correctly, Elliotte Rusty Harold was asking for that, 
but Norm wasn't. Norm, was that something that you support?

At any rate, the current mappings do not support *any* W3C XML-Schema 
semantics in the Infoset-only mapping. I do think this is a slippery slope, 
and I wouldn't support it. In effect, both you and Elliotte are asking that 
we support the syntax of XML 1.x, but change the semantics by adding some 
pieces that each of you like.

> > On the other hand, we give plenty of flexibility for an implementation to
> > create data model instances without using our PSVI or Infoset mappings.
>
>I acknowledge that, but I want something that implementations will provide,
>by making it a standard part of XPath processing, rather than something an
>idiosyncratic implementation might provide.

But it's not just XPath, is it? Suppose the path expression were issued 
from the DOM, wouldn't you like those types to be available in the DOM 
representation as well? Personally, I would like to see a small, simple way 
to add datatypes to documents governed by DTDs, but I would also like to 
see it be part of general XML processing, not something used to doctor up 
documents just before they are queried, and not recognized by the rest of 
my XML tools.

I think that XPath should operate on the data given it by the XML specs. 
What you want makes sense, but this is the wrong place in the processing 
stack to try to do it in a standard way.

Jonathan

Received on Thursday, 5 June 2003 15:19:25 UTC